Motion Diffusion Autoencoders: Enabling Attribute Manipulation in Human Motion Demonstrated on Karate Techniques

TL;DR

This paper introduces a novel method for attribute manipulation in human motion data, specifically karate movements, using a combination of a new pose representation, transformer encoders, and diffusion models to achieve controllable and meaningful motion editing.

Contribution

It presents the first successful approach for attribute manipulation in human motion, utilizing a new pose representation and a hybrid model of transformers and diffusion processes.

Findings

Semantic embedding space is linear and meaningful.

Attribute manipulation is achieved by moving embeddings along linear directions.

Method enables precise control over human motion attributes.

Abstract

Attribute manipulation deals with the problem of changing individual attributes of a data point or a time series, while leaving all other aspects unaffected. This work focuses on the domain of human motion, more precisely karate movement patterns. To the best of our knowledge, it presents the first success at manipulating attributes of human motion data. One of the key requirements for achieving attribute manipulation on human motion is a suitable pose representation. Therefore, we design a novel continuous, rotation-based pose representation that enables the disentanglement of the human skeleton and the motion trajectory, while still allowing an accurate reconstruction of the original anatomy. The core idea of the manipulation approach is to use a transformer encoder for discovering high-level semantics, and a diffusion probabilistic model for modeling the remaining stochastic variations. We show that the embedding space obtained from the transformer encoder is semantically meaningful and linear. This enables the manipulation of high-level attributes, by discovering their linear direction of change in the semantic embedding space and moving the embedding along said direction. All code and data is made publicly available.